CPT-Vs correlations for post-glacial sediments in Metropolitan Vancouver

Abstract

A comprehensive Seismic Cone Penetration Test (SCPT) database for post-glacial sediments in Metro Vancouver is compiled to assess and establish relationships between CPT parameters and shear-wave velocity (Vs). Five world-wide and one existing Fraser River Delta (FRD) specific empirical CPT-Vs correlations are examined and found to have limited applicability in predicting Vs from our SCPT database. Region-specific CPT-VS models are then developed from a training dataset using 4 different regression approaches and 6 different CPT parameters combinations. The regression approaches include multi-linear (MLR) and nonlinear (NLR) regressions, and Random Forest and Extreme Gradient Boosting supervised ensemble machine learning models. The performance of different models and CPT parameters combinations is assessed using a testing dataset. The machine learning models are found to perform slightly better than MLR and NLR models in predicting Vs. Based on performance and practicality, a final CPT-Vs model is selected and applied to predict Vs for 59 compiled CPTs in the region. The comparison between predicted CPT-Vs profiles and measured Vs profiles in similar geology units confirms accuracy of our developed model. The developed CPT-Vs prediction models can be employed to obtain Vs profiles for site response analysis, compare CPT- and Vs-based liquefaction evaluation methods, or map Vs30 in Metro Vancouver.